Grid Tagging Schemes as Semantic Segmentation

• Published in: 2024 7th International Conference on Information and Computer Technologies (ICICT) pp. 369 - 376
• Main Author: Qi, Zhengyu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.03.2024
• Subjects: Benchmark testing; Error analysis; Feature extraction; Lattices; Natural Language Processing; OFOE; Semantic segmentation; Sentiment Analysis; Tagging; Transforms
• ISSN: 2769-4542
• DOI: 10.1109/ICICT62343.2024.00066

Oriented Fine-grained Opinion Extraction (OFOE) is a crucial task in Natural Language Processing (NLP) aimed at the automated extraction of aspect and opinion terms from textual comments. This extraction process typically results in the formation of aspect-opinion pairs, or in a more advanced form, aspect-opinion-sentiment ternaries that also encapsulate sentiment polarity. A contemporary methodology, the Grid Tagging Scheme (GTS), has demonstrated its efficacy in managing OFOE tasks through an end-to-end approach. Building upon this foundation, our research introduces a novel methodology that reconceptualizes the GTS framework as a task of semantic segmentation. This innovative approach entails performing semantic segmentation on both the opinion pair extraction (OPE) and the opinion triplet extraction (OTE) lattices, derived from the GTS. These lattices undergo labeling before the initiation of inference operations. The distinctive feature of this approach lies in its ability to assimilate both local and global contextual information, leading to varied levels of precision, recall, and F1-score across four benchmark datasets, in contrast to the results achieved by the GTS. This difference underscores the potential of our approach in enhancing the accuracy and efficiency of OFOE tasks.